Trace Feedback

Use Feedback to tie human judgment to a specific trace so you can validate fixes and catch regressions early. It turns "this felt wrong" into a trackable signal you can compare across runs.

What to look at:

• History and source: confirm the trace was rated and where it came from (app, API, model).
• Key and score: keep signals consistent (for example, satisfaction) so you can compare runs.
• Comments: capture short context that explains why the output was good or bad.